Posterior Model Probabilities after RoBMA

[bobreednz]

JASP Version

0.95.1

Commit ID

No response

JASP Module

Meta Analysis

What analysis are you seeing the problem on?

summary(fit, type = "models")

What OS are you seeing the problem on?

Windows 11

Bug Description

It is my understanding that the command "summary(fit, type = "models")" should produce, among other things, posterior model probabilities. But it doesn't (see attached). I used the Bem2011 data. The program was taken from the "Show RoBMA R code" under "Advanced". I have copied that code below.

fit <- RoBMA(
d = dataset[['d']],
se = dataset[['se']],
prior_scale = 'cohens_d',
transformation = 'fishers_z',
priors_effect = list(
prior(distribution = 'normal', parameters = list(mean = 0, sd = 1), truncation = list(lower = -Inf, upper = Inf), prior_weights = 1)
),
priors_heterogeneity = list(
prior(distribution = 'invgamma', parameters = list(shape = 1, scale = 0.15), truncation = list(lower = 0, upper = Inf), prior_weights = 1)
),
priors_effect_null = list(
prior(distribution = 'point', parameters = list(location = 0), prior_weights = 1)
),
priors_heterogeneity_null = list(
prior(distribution = 'point', parameters = list(location = 0), prior_weights = 1)
),
priors_bias = list(
prior_weightfunction(distribution = 'two.sided', parameters = list(steps = 0.05, alpha = c(1, 1)), prior_weights = 0.0833333333333333),
prior_weightfunction(distribution = 'two.sided', parameters = list(steps = c(0.05, 0.1), alpha = c(1, 1, 1)), prior_weights = 0.0833333333333333),
prior_weightfunction(distribution = 'one.sided', parameters = list(steps = 0.05, alpha = c(1, 1)), prior_weights = 0.0833333333333333),
prior_weightfunction(distribution = 'one.sided', parameters = list(steps = c(0.025, 0.05), alpha = c(1, 1, 1)), prior_weights = 0.0833333333333333),
prior_weightfunction(distribution = 'one.sided', parameters = list(steps = c(0.05, 0.5), alpha = c(1, 1, 1)), prior_weights = 0.0833333333333333),
prior_weightfunction(distribution = 'one.sided', parameters = list(steps = c(0.025, 0.05, 0.5), alpha = c(1, 1, 1, 1)), prior_weights = 0.0833333333333333),
prior_PET(distribution = 't', parameters = list(location = 0, scale = 1, df = 1), truncation = list(lower = 0, upper = Inf), prior_weights = 0.25),
prior_PEESE(distribution = 't', parameters = list(location = 0, scale = 5, df = 1), truncation = list(lower = 0, upper = Inf), prior_weights = 0.25)
),
priors_bias_null = list(
prior_none(prior_weights = 1)
),
effect_direction = if (median(dataset[['d']]) >= 0) 'positive' else 'negative',
chains = 3,
adapt = 3500,
burnin = 4500,
sample = 10000,
thin = 1,
autofit = FALSE,
algorithm = 'ss'
)

Expected Behaviour

I would have expected to see prior and posterior model weights.

Steps to Reproduce

Reran the program.

Log (if any)

No response

More Debug Information

No response

Final Checklist

• I have included a screenshot showcasing the issue, if possible.
• I have included a JASP file (zipped) or data file that causes the crash/bug, if applicable.
• I have accurately described the bug, and steps to reproduce it.

[FBartos]

Hi @bobreednz (Bob?)

This issue is more relevant for the RoBMA package itself. I created the corresponding feature request there (FBartos/RoBMA#72) and I will just quickly summarize and close the issue here.

The summary(fit, type = "models") call (currently) produces posterior model probabilities only for the bridge-sampling algorithm. The algorithm = "ss" version presently used in JASP (which is necessary for faster model fitting in meta-regression / hierarchical models) does not fit the models separately. As such, it does not produce a model-wise summary by default -- I can add a feature that adds this summary based on the posterior samples.

Cheers,
Frantisek